Decoding PFAS contamination via Raman spectroscopy: A combined DFT and machine learning investigation

In this study, density function theory (DFT) is employed to compute Raman spectra of 40 important Perfluoroalkyl substances (PFASs) as listed in Draft Method 1633 by U.S. Environmental Protection Agent. A systematic comparison of their spectral features is conducted, and Raman peaks and vibrational modes are identified. The Raman spectral regions for the main chemical bonds (such as C-C, CF2 & CF3, O-H) and main functional groups (such as -COOH, -SO3H, -C2H4SO3H, and -SO2NH2) are identified and compared. The impacts of branching location in isomer, molecular chain length, and functional groups on the Raman spectra are analyzed. Particularly, the isomers of PFOA alter the peak locations slightly in wavenumber regions of 200 – 800 and 1000 – 1400 cm−1, while for PFOS, spectral features in the 230 – 360, 470 – 680, and 1030 – 1290 cm−1 regions exhibit significant difference. The carbon chain length can significantly increase the number of Raman peaks, while different functional groups give significantly different peak locations. To facilitate differentiation, a spectral database is constructed by introducing controlled noise into the DFT-computed Raman spectra. Subsequently, two chemometric techniques, principal component analysis (PCA) and t-distributed stochastic neighbor embedding (t-SNE), are applied to effectively distinguish among these spectra, both for 40 PFAS compounds and the isomers. The findings demonstrate the promising potential of combining Raman spectroscopy with advanced spectral analysis methods to discriminate between distinct PFAS compounds, holding significant implications for improved PFAS detection and characterization methodologies.